Non-independence in statistical tests for discrete cross-species data

Area of Science:

• Phylogenetics
• Statistical methods
• Evolutionary biology

Background:

• Statistical tests for character state associations in cross-species data are crucial for evolutionary inference.
• Existing methods may be susceptible to biases arising from data structure and reconstruction techniques.
• Understanding these biases is essential for accurate phylogenetic analyses.

Purpose of the Study:

• To identify and describe previously undetected biases in statistical tests for character state associations.
• To investigate the 'family problem' and its interaction with non-independence in character state reconstruction.
• To evaluate the impact of these biases on phylogenetic data analysis.

Main Methods:

• The study identifies biases arising from non-independence in statistical tests.
• It introduces the 'family problem,' a general bias affecting character state reconstruction.
• It analyzes biases specific to joint and single character state reconstruction methods.

Main Results:

• Three previously undetected biases in statistical tests for character state associations were identified.
• The 'family problem' is a general bias affecting all tested methods.
• Biases in character reconstruction methods interact with the family problem, influencing results.

Conclusions:

• The family problem significantly impacts the reliability of statistical tests for character state associations.
• The choice of character reconstruction method influences the type and magnitude of bias observed.
• Accurate phylogenetic inference requires careful consideration of these identified biases and their interactions.